Project Overview

Day 1

  • Plot the value function against the state variables. What is the intuition?
  • Plot the policy function against the state variables. What is the intuition?
  • How do the two plots above change with the values of \(\alpha\) and \(\delta\)?
  • Write a subroutine that takes the model’s solution and parameters as inputs, then produces a simulated data set as output.
  • Use the Tauchen (1986) method to discretize the state space.

Day 2

  • Which features of the data will you ask the model to fit? These features may include means, variances, regression coefficients, or correlations.
  • Explain which moment or moments are most important for identifying each model parameter. Create a table showing how each simulated moment changes as you perturb \(\alpha\) and \(\delta\).
  • Downlad the data and read the documentation. The data set contains cleaned Compustat data on non-financial firms from 1971 - 2016. These variables should give you a hint about the types of moments that will work well.
  • Write a subroutine for computing the vector of moments from the actual data and simulated data.
  • If one of your moments is an AR1 coefficient, use the Han and Philips method at the end of Toni’s slides from today.

Day 3

  • Use influence functions to estimate the covariance matrix for the moments you have chosen, clustering by firm.
  • Compute the optimal SMM weight matrix.
  • Code up the SMM routine.
  • Write the code for finding the parameter vector that minimizes the econometric score, i.e. the distance between the actual and simulated moments.
  • Choose a minimizer that will avoid local minima, such as the simulated annealing minimization algorithm or particle swarm optimization.

Day 4

  • SMM estimator
  • Compute your final parameter estimates and their standard errors. Do they make sense?
  • How well does the model fit the data? Compare the simulated and empirical moments.
  • Does the model fail the test of overidnentifying restrictions?
  • What would be an interesting counterfactual experiment to run? How would you do it?